If I am trying to submit more remote jobs then available resources, then my understanding is that those remote jobs are queued up for execution, correct?
What is the practical limit for the submission queue? Do I have any visibility into it? Is there an option to somehow persist it, if the cluster goes down and restarts? In other words, what are the guarantees that Ray provides on such queue?
Also, once submitting a a remote job, can I disconnect and come back later to see the state in the queue and results, if the job has executed?
These are great questions – I think the linked thread is a good place to discuss things on the theme of job queues: